WO2004099985A1

WO2004099985A1 - Execution environment danger prediction/evading method, system, program, and recording medium thereof

Info

Publication number: WO2004099985A1
Application number: PCT/JP2003/005814
Authority: WO
Inventors: Yukihiro Kimura; Masakazu Hayashi; Katsutomo Sekiguchi
Original assignee: Fujitsu Limited
Priority date: 2003-05-09
Filing date: 2003-05-09
Publication date: 2004-11-18
Also published as: JPWO2004099985A1; US20050240641A1; JP4170988B2; US7516292B2

Abstract

Measurement/analysis means (10) in a Java virtual machine (1) includes a function for measuring data when a garbage collection event occurs and a function for periodically measuring data. The measurement/analysis means (10) analyzes the measured data and predicts danger of memory shortage. Here, danger prediction is performed by the algorithm corresponding to the garbage collection type used by the Java virtual machine (1). The alarm of danger of memory shortage predicted is notified to an application server (2) by analysis result notification means (31) such as inter-process communication. Moreover, the measurement/analysis means (10) calculates a memory amount required for evading the danger of memory shortage predicted. The calculated memory amount is notified to the application server (2) by the analysis result notification means (31) such as the inter-process communication in the same way as the alarm.

Description

Execution environment danger prediction / avoidance method, system, program and its recording medium

The present invention relates to a technology for stably operating an application program, and in particular, a client layer and a middle layer constituting an enterprise application.

In a three-tier system with a database layer, the execution environment of the middle tier, which is located on the server side, is monitored during the operation of the application to prevent the execution environment from becoming dangerous, thereby stabilizing the application. Forecasting the danger of the execution environment for operating the system.

Examples of the enterprise applications include a Web application mainly using Java (registered trademark).

In the configuration of the enterprise application described above, there is also a four-layer system in which the middle layer is divided into two layers.

The execution environment of the middle tier located on the server side is mainly a Java Virtual Machine (Java Virtual Machine).

The present invention is intended for an execution system having a memory management function such as garbage collection such as JaVa. Background art

Hereinafter, J a v a will be used as an example for explanation.

Fig. 1 is a diagram showing an example of the operating environment of a Java application. In the example of Fig. 1, data during the operation of the application in the execution environment 100 of the JaVa application is collected, and the collected data is analyzed by the performance analysis tool 2000. I do. By feeding back the analysis results of the performance analysis tool 2000 to the execution environment 100 of the JaVa application, an optimal operation environment for the Java application is constructed.

In the execution environment of Java application 100, higher than OS 110 In the layer, there is a Java virtual machine 100, and in the upper layer, the application server

There is 1 0 3 0. As an example of the application server 1,030, there is an EJB container ゃ Serv1et container which is a program for operating EJB (Enterprise JavaBeans) and Serv1et. In the upper layer of the application server 10030, applications 104 such as Servleet, Java Server Pages (JSP), and EJB are operated. The Web server 1050 accepts a request from the client. The most familiar with the operation status of the J aV a application is the J a V a virtual machine 1 0 0 which is located at the lowest level in the execution environment 1 It is.

In the performance analysis tool 2000, data collected from the execution environment 1000 is stored in the performance database 210, and the stored data is stored in the analysis program 200.

Parsed by 20. Display the analysis results with the monitoring tool 203

, Use the drill-down tool 204 for more detailed analysis. In addition, the service level diagnosis 250 diagnoses whether the server responds to the request from the client within the specified time or whether the current service status is satisfied.

Figure 2 shows an example of a conventional method for analyzing the operating environment of a Java application. In Fig. 2, the data collected from the server 300 running the Java application (the server having the Java application execution environment 100) is stored in the performance database 210. It is stored in the data storage device 31010 provided. The performance analyzer 320 refers to the data stored in the data storage device 310, and analyzes the data using the analysis program 220. The operation administrator 3003 can use the performance analyzer 30020 to display analysis results in a graph using the monitoring tool 2300, perform detailed analysis using the drilldown tool 204, The following technologies are known as conventional technologies related to stable operation of enterprise applications that perform service level diagnosis by using service level diagnosis 20000.

(First conventional technology) There is a technology in which the operation manager monitors the processing performance of services and the system performance of the server side using monitoring tools. This technology analyzes the measured data, predicts and detects signs of a trap in conjunction with the threshold, and notifies the operation manager of it. In addition, by analyzing the measured data with an analysis tool, the operation status can be graphed and evaluated, and detailed analysis of performance data (drill-down) can be performed to support the identification of bottlenecks in performance degradation. The example in Fig. 2 is an example of this first prior art. The processing performance of the above service is how many seconds it can respond to a request from a client. The server-side system performance described above includes CPU usage, memory usage, and network usage.

The following documents describe the first conventional technology.

Non-patent document 1

Systemw alker, Performance Management Products, Fujitsu Limited, (URL: http: // systemwalker. Fujitsu. Com / jp / com / conc / pafo.html)

Non-patent document 2

System Management / Tivo 1 i, Performance & Availability Products, Japan IBM Co., Ltd. (UR L: http://www-6.ibm.com/jp/software/tivoli/products/inaex.html#per)

Non-patent document 3

JEEE, JGC Information Software, Inc., (URL: http: //www.jsys-products.com/product/pre_md_Zee/document, html)

(2nd conventional technology)

In an environment where servers with different processing performance coexist, the processing time of services such as Servlet ゃ EJB is measured, the server with the highest processing speed is automatically selected, and requests from clients are distributed to efficiently process the entire system. There is a technology to perform.

The following documents describe the second conventional technology.

Non-patent document 4 I BM e Server, Heterogeneous Workload Management, Japan I BM Co., Ltd., U "RL: http: //www-6.ibm.com/jp/servers/eserver/ac/tech/index2-1 html)-(3rd prior art)

There is a technology that detects the memory shortage by periodically examining the memory (heap) size managed by the Java virtual machine using an application program interface (API) that checks the amount of Java memory. In this technology, when the free space of the memory is less than the threshold value, the garbage collection is forcibly executed to encourage the recovery of the free memory.

The following documents describe the third conventional technology.

Non-patent document 5

Web Logic Server, Java Virtual, machine tuning, automatic detection of low memory status and forced garbage collection, Japan BEA Systems Corporation, (URL: http://edocs.beasys.co.jp / e-docs / wls / docs70 / perform / JVMTuning.html # 489159)

(4th conventional technology)

Java virtual machine · The profiler interface measures events generated in the Java virtual machine, detects long-running methods (bottlenecks), detects objects with memory leaks, and risks deadlocks. There is a technology that detects a possible thread.

Events that are measured by the Java Virtual Machine · Profiler interface include "garbage collection started and finished", "Java method entered and finished", "object allocated / released", " The status of the Java thread has changed. "

However, the conventional technologies for stable operation of enterprise applications as described above have the following problems.

(Problem of the first prior art)

In the first conventional technique, a monitoring tool that collects various data, displays it in a graph, etc. and monitors it requires human intervention in the steps of trouble prediction and avoidance. In short, it lacked timeliness and autonomy. In particular, there was no timeliness in the method of analysis using a drill-down tool. In addition, even though data collection, service level diagnosis (analysis), and notification of warnings were performed automatically, many measures were not automated after receiving a warning. In addition, conditions such as how much data is collected and when analysis is performed have affected the timing of danger discovery.

(Problem of the second conventional technology)

In the second conventional technology, the processing is performed automatically. However, this is a load balancing mechanism that measures the service time and distributes the load. Was not a technology to discover.

(Problem of the third conventional technology)

In the third conventional technology, processing is performed automatically, but there is a problem with the algorithm for predicting trouble. Memory (heap) lack of J _a V _a virtual machine is the main cause of tiger pull, but it can not be said that it is always right decision only detect-heap by comparing the available capacity and the threshold value.

When the heap becomes insufficient, the JaVa virtual machine automatically executes garbage collection to recover memory. However, in the third conventional technique, a forced garbage collection is performed before that. We are trying to recover memory. This technique may be useful if the memory usage gradually increases after the memory usage exceeds the threshold and it takes a relatively long time for automatic garbage collection to occur.

However, not all situations are such. For example, if the absolute amount of memory is insufficient and the frequency of garbage collection is high, the next garbage collection will occur immediately after the forced garbage collection occurs. I will go again.

(4th problem of the prior art)

In the fourth prior art, the events provided by the JaVa virtual machine / profiler interface are issued very frequently, which places an extra load on the JaVa virtual machine. In addition, a predetermined interface As a result, they did not have the information they needed to find the danger, or they had a low degree of freedom. Many tools that find sequence calls (bottlenecks) or memory leaks with high CPU usage use a profiler interface, but the load is relatively high, and these are not used during application operation but during the testing stage. Was to be done.

In order to solve the above problems of the conventional technology and realize a mission-critical system by operating applications stably, it is necessary to detect and avoid the danger of trouble before an error occurs. To do so, the following conditions must be met from the discovery of the danger to its avoidance.

(1) Reduce the load of data collection for predicting traps, and do not impose a load on application operation.

(2) Use appropriate judgment materials and algorithms for predicting troubles, and make more accurate predictions. .

(3) Perform processing from data measurement to predicting, notifying, and avoiding problems in a short time and in a timely manner. In addition, all of these steps should be performed automatically (autonomously).

The present invention has been made in view of such a situation, and reduces the load of data collection, performs more accurate risk prediction from appropriate judgment materials and algorithms, and predicts trouble from data measurement. The purpose of the present invention is to provide a technology that can perform processing up to notification, avoidance, and avoidance in a short time and in a timely manner, and can automatically execute the processing. Disclosure of the invention

The present invention is, for example, a method for predicting and avoiding the danger of memory shortage in an execution environment of a middle tier of an enterprise application, and a method for estimating and avoiding the danger of the execution environment to avoid the danger. This danger prediction / avoidance method measures the data related to the memory status when a garbage collection event occurs, predicts the danger of insufficient memory from the measured data, and issues a warning about the predicted danger of insufficient memory. Notify other applications. It also avoids the predicted danger of running out of memory Calculate the amount of memory required for this and notify the calculated amount of memory. In addition, prediction of the danger of memory shortage is performed using algorithms that correspond to each type of garbage collection.

Specifically, for example, when a Java virtual machine is used as the execution environment of the middle tier of an enterprise application, when an event of garbage collection occurs, data such as the time of occurrence / memory usage is measured. Other than the above, the memory usage is regularly measured. Based on the measured data, it predicts the danger of a memory shortage error (hereinafter referred to as OutOfMemory) or performance degradation due to memory shortage, and notifies them of such warnings. Also, calculate the amount of memory required to avoid danger. To predict the danger of memory shortage, predictions are made not only for basic garbage collection but also for special garbage collection such as generational garbage collection.

According to the method for avoiding danger prediction of the execution environment according to the present invention, when an event other than garbage collection occurs (for example, when an event of memory allocation occurs), the memory usage is not measured. Since the measurement of memory usage is performed by measurement at the time of event occurrence and periodic measurement at a relatively long cycle as necessary, data can be measured without imposing a load on application operation. it can. Also, according to the method for avoiding danger prediction of the execution environment according to the present invention, the amount of memory required to avoid the predicted danger of memory shortage is calculated, and the calculated amount of memory is notified. The execution environment can be quickly and easily recovered. In addition, according to the method for estimating and avoiding danger of the execution environment according to the present invention, danger is determined by an algorithm corresponding to each type of garbage collection, so that a more accurate danger can be determined. The method for avoiding danger prediction of the execution environment according to the present invention can be realized by a computer and a software program, and the program can be recorded on a computer-readable recording medium or can be provided through a network. It is. This program is installed on a medium such as a hard disk, expanded in main storage, and executed by the CPU. In addition, recording media Portable recording media such as CD-ROMs and DVDs are conceivable. BRIEF DESCRIPTION OF THE FIGURES

Fig. 1 is a diagram showing an example of the operating environment of the Java application.

Figure 2 shows an example of a conventional method for analyzing the operating environment of a Java application.

Fig. 3 is a diagram showing an example of the configuration of the danger prediction / avoidance system.

FIG. 4 is a diagram showing an example of setting data.

Fig. 5 is a diagram for explaining the measurement and analysis means.

Fig. 6 is a diagram showing three typical states of garbage collection intervals and memory usage.

FIG. 7 is a diagram illustrating a state in which a warning should be given.

Fig. 8 is a diagram explaining the garbage collection interval and memory usage in generational garbage collection.

Fig. 9 is a diagram explaining the measurement of the overflow amount.

Fig. 10 is a diagram for explaining the state of garbage collection intervals and memory usage in generational garbage collection that dynamically changes size.

Figures 11 to 15 are flowcharts of the memory usage measurement and analysis process.

Fig. 16 and Fig. 17 show examples of the status of the memory usage measurement and analysis means that is the target of warning.

FIG. 18 is a flowchart of the thread measurement / analysis processing.

Fig. 19 is a diagram showing the distribution of processing to the Java virtual machine by cooperation between the Java virtual machine and the application server.

Figure 20 is a diagram showing the cooperation between the Java virtual machine and the application server.

FIG. 21 is an analysis result notification processing flowchart.

FIG. 22 is a flowchart of a command notification process. FIG. 23 is a flowchart of the JaVa virtual machine optimal control process. BEST MODE FOR CARRYING OUT THE INVENTION

Fig. 3 is a diagram showing an example of the configuration of the danger prediction / avoidance system. In FIG. 3, the Java virtual machine 1 and the application server 2 in the upper layer are linked by the Java virtual machine linking means 3.

The Java virtual machine 1 is provided with measurement and analysis means 10. The measurement / analysis means 10 includes a memory usage measurement / analysis means 11, a thread measurement / analysis means 12, and setting data 100.

The application server 2 includes a Java Virtual Machine control means 20. Also, the Java virtual machine control means 20 includes a Java virtual machine optimum control means 21.

The JaVa virtual machine cooperation means 3 includes an analysis result notifying means 31 and a command notifying means 32.

FIG. 4 is a diagram showing an example of setting data. As shown in Fig. 4, the setting data 100 provided by the measurement and analysis means 10 is the memory usage measurement flag 101, the thread measurement flag 102, the GC generation interval warning threshold (thl) 103 , OutOfMemory warning threshold A (th 2) 104, OutOfMemory warning threshold B (th 3) 1 0 5, Out of memory warning threshold (th 4) 1 06, Memory free recovery threshold (th 5) 107, new generation shortage warning threshold (th6) 108, overflow warning threshold (th7) 109, deadlock warning threshold (th8) 110. The above GC means garbage collection. Hereinafter, garbage collection is sometimes referred to as GC.

These values in the setting data 100 can be set by the application server 2 issuing a command. In addition, a system administrator or the like can set in advance or at any time using a setting program (not shown).

The operation of each means is described below. First, the part that measures, analyzes, and predicts the danger of data in the JaVa virtual machine 1 will be described. And feed packs for optimal operation. Will be described.

In the present invention, as shown in FIG. 3, the JaVa virtual machine 1 includes a measurement-analysis means 10 therein. Measurement and analysis of data and prediction of the occurrence of traps are performed inside the Java virtual machine 1. Therefore, in the present invention, the process from data measurement to analysis and trouble prediction is performed in a more timely manner than in the first conventional technology in which measured data is output once to a database or the like and judged by an analysis program. be able to.

Fig. 5 illustrates the measurement and analysis means. Measurement inside the J aV a virtual machine 1-The analysis means 10 is a function that periodically measures data (for example, at intervals of several seconds) 10 α and a function that measures data when a garbage collection event occurs 1 It has two data measurement functions of 0 β.

The exact amount of memory (heap) used in the JaVa virtual machine 1 is measured at the time of the "memory release" event. However, the frequency of events is extremely high, and measuring data when all events occur places a burden on application operations. Analyzing memory usage and predicting danger can be performed on data that is acquired periodically, and there is no need to measure data when a “memory allocation Z release” event occurs.

Measurement 'Analysis means 10 is used from two data measurement functions (10α, 10j3) according to the items to be measured so that data can be accurately measured without imposing a load on application operation. Select the data measurement function.

As shown in Fig. 3, the measurement inside the JaVa virtual machine 1 is performed by the analysis means 10 and the memory usage measurement and analysis means 11 for measuring and analyzing the memory usage. Thread measurement and analysis means 12 for measuring and analyzing the thread are provided.

First, the memory usage measurement and analysis means 11 will be described. Memory usage measurement • Analytical method 11 1 periodically measures the amount of used memory (heap) and measures the occurrence of garbage collection in the JaVa virtual machine 1, and obtains the results of those measurements. Analyze the data. As a result of the analysis, if there is a danger that OutMemory errors may occur due to insufficient memory, or if insufficient memory adversely affects performance, a warning is given. . It is also necessary to restore the status at that time. Estimate the moly size.

Memory usage measurement · Analysis unit 11 measures the amount of memory used by Java virtual machine 1 at regular intervals. The setting of the measurement interval at this time can be changed by a command from the & ゃ virtual machine control means 20 (the details will be described later) of the option 1 when starting the && virtual machine 1 and the application server 2 It is. In addition, the memory usage measurement 'analysis means 11 measures data when a garbage collection event occurs.

] _& 3 Since the virtual machine 1 knows its own operation status well, it can detect the risk of trouble within itself. The most serious problem in the JaVa virtual machine 1 is a lack of memory (heap). Due to this memory shortage, machine performance is degraded, and in severe cases, processing may stop.

Also, the amount of transaction requests from clients may be larger than expected, and the memory usage may exceed the initial estimate. The amount of requests from clients changes every moment, and the amount of memory used also changes according to the situation. In such a case, it is not possible to accurately judge the danger of memory shortage simply by comparing the amount of used memory measured with the threshold value as in the third conventional technique described above. Therefore, it is necessary to make more appropriate decisions according to the situation. '

In the application of JaVa, when memory is required, the memory is acquired by the new operator. In a Java application, memory release cannot be written in a program, and memory release is left to the Java virtual machine 1. The process of releasing memory in JaVa virtual machine 1, that is, garbage collection, has a very large effect on machine performance. Therefore, various types of garbage collection according to the situation are being studied, and the garbage collection method adopted by each Java Virtual Machine 1 is different.

Therefore, in order to determine the danger of running out of memory, in addition to the common determination method that can be applied to all garbage collections, the determination method specific to the garbage collection used for each JaVa virtual machine 1 Required. Memory usage measurement and analysis unit 1 1 analyzes the measured data, and determining the risk of a common memory shortage applicable to all gas page collection, is adopted for each J _a V a virtual machine 1 Determines the risk of running out of memory inherent to garbage collection. Also, based on the results of each judgment, a warning is issued for a dangerous state and the amount of memory required to recover the state is calculated.

Measurement of memory usage · Analysis of basic data by means of analysis 11, warning for dangerous conditions, and calculation of the amount of memory required to recover the status are described. First, the three typical states of the garbage collection interval and the memory usage are explained using figures.

Fig. 6 is a diagram showing three typical states of the garbage collection interval and the memory usage. The graphs shown in Fig. 6 (A) to (C) are examples of graphs showing changes in memory usage 200 and the occurrence of garbage collection 210, and the occurrence of garbage collection 210. Three typical states related to intervals and memory usage 200 are shown.

In the graphs in Figs. 6 (A) to 6 (C), the vertical axis indicates the amount of memory, and the horizontal axis indicates time. In Fig. 6 (A) to (C), the solid line shows the time change of the memory usage 200, and the dotted line shows the set maximum heap size 220. JaVa virtual machine 1 executes garbage collection 210 when the memory usage 200 is likely to exceed the maximum heap size 220 due to the "memory allocation" event, and releases the memory. Do.

Figure 6 (A) shows a stable state where garbage collection 210 does not occur, or the interval between garbage collection 210 occurrences is relatively long, and garbage collection 210 recovers memory space. Indicates the state. Fig. 6 (B). Shows the state where the interval of garbage collection 210 tends to be short, but the memory space is recovered by garbage collection 210. Fig. 6

(C) shows a state in which the garbage collection 210 occurs at short intervals and the garbage collection 210 does not sufficiently recover the memory space. The state shown in Fig. 6 (B) and the state shown in Fig. 6 (C) are subject to warning.

FIG. 7 is a diagram illustrating a state in which a warning should be given. Figure 7 (A) shows the state in which a warning about performance degradation due to insufficient memory should be issued. In Fig. 7 (A), the vertical axis indicates the intervals of garbage collection, and the horizontal axis indicates the intervals of garbage collection at equal intervals. In this example, the garbage collection occurrence interval has fallen below the GC occurrence interval warning threshold 103.

Figures 7 (B) and 7 (C) are diagrams showing the state in which a warning of the danger of OutOfMemory should be issued. In Figs. 7 (B) and 7 (C), the vertical axis shows the amount of memory and the horizontal axis shows time. In Fig. 7 (B), the bold line 104 is the value obtained by (full capacity XOutOfMemor y warning threshold A 104). In Fig. 7 (C), the bold line 105 'is the value obtained by (total capacity X OutOfMemory warning threshold B105).

Looking at the change in the memory usage 200 in Fig. 7 (B), almost no free memory has been recovered by GC210. Since the memory usage 200 after GC 210 exceeds the thick line 104 ', it can be seen that the memory usage rate (memory usage Z total capacity) exceeds the OutOfMemory warning threshold A104. .

In Fig. 7 (C), a change in the memory usage of 200 indicates that the free space of the memory is gradually recovering in the order of GC210a, GC210b, and GC210c. . In addition, since the memory usage 200 after GC 210b and GC 210c exceeds the thick line 105 ', the memory usage after GC 210b and GC 210c is used. It can be seen that the rate is above the OutOfMemory warning threshold B105.

Here, it is assumed that the OutOfMemory warning threshold value A 104 is larger than the OutOfMemory warning threshold value B 105. Even if the memory usage is below the OutOfMemory warning threshold A104, if the memory usage is increasing above the OutOfMemory warning threshold B105, there is a danger of OutOfMemory. The following (Judgment 1) to (Judgment 3) show the tricks to judge the risk of memory shortage in the basic garbage collection by the memory usage measurement 'analysis means 11'. In the following basic garbage collection judgment, the usage rate is a memo. Indicates the ratio of the memory usage to the total capacity of the memory. Therefore, the following equation (1) is obtained.

Usage rate = memory usage / total capacity · · · formula (1)

(Decision 1) The garbage collection interval is measured, and if the interval continues below the GC interval interval warning threshold of 103, it is considered that performance is degraded due to insufficient memory. Judge (see Fig. 7 (A)).

(Judgment 2) The memory usage after garbage collection is measured, and if the usage rate exceeds the OutOfMemory warning threshold A104, it is judged that there is a danger of OutOfMemory (Fig. 7 ( B))).

(Decision 3) The memory usage after garbage collection is measured, and if the usage rate is continuously increasing and exceeds the power S Out Of Memory warning threshold B105, the OutOfMemory Judge that there is a danger (see Fig. 7 (C)).

Memory usage measurement and analysis means 11 Based on the judgment results of basic garbage collection dangers shown in (Judgment 1) to (Judgment 3), warning of OutOfMemory danger, memory shortage Warning of performance degradation due to

The memory usage measurement and analysis means 11 calculates the amount of memory required to recover the state. For example, the memory size required for recovery in (Judgment 1) to (Judgment 3) is such that the vacancy rate of the memory after garbage collection (vacancy rate = 1—usage rate) exceeds the memory vacancy recovery threshold value of 107. Is the size calculated in The calculation is to find m (capacity added for recovery) that satisfies the following equation (2).

(Free space + m) / (Total space + m)

> Memory free recovery threshold value 1 0 7 ··· 'Equation (2) or more, memory usage measurement · Basic data analysis by analysis means 11, warning for dangerous status, amount of memory required for status recovery Has been described. Furthermore, by adding and improving judgments according to the characteristics of each garbage collection, it is possible to analyze data more accurately, warn of dangerous conditions, and calculate the amount of memory required for state recovery.

Next, the memory usage measurement and analysis means 11 analyzes the data according to the characteristics of the generational garbage collection, warns against dangerous conditions, and provides information necessary to recover the status. The calculation of the amount of moly will be described.

Generation-type garbage collection manages the memory (heap) by dividing it into two areas, the new generation and the old generation. When a Java program object is allocated to memory, it is first allocated to a new generation of memory. When the new generation of memory becomes full, a new generation of garbage collection (hereafter, newGC) will occur. Objects that have survived n e wGC several times are moved to the old generation memory. When the old generation memory becomes too old, garbage collection of the entire memory (hereinafter simply referred to as garbage collection or GC) occurs. Therefore, the memory usage of the new generation will be very volatile.

Among the APIs of JaVa shown in the third prior art mentioned above, there is an API that calculates the total memory capacity and the free memory capacity. However, if these APIs return the total memory size of the new generation and the old generation as data, the graph shown in Fig. 6 does not become the graph even if the data is graphed. Therefore, the data cannot be analyzed correctly. In the case of generational garbage collection, it is necessary to analyze the data by separately measuring the memory usage of the new generation and the old generation.

Fig. 8 is a diagram explaining the state of the garbage collection interval and the memory usage in the generation type garbage collection. The graphs shown in Figs. 8 (A) to (C) show the changes in memory usage (old generation 201, new generation 202) and garbage collection (GC) 210, new GC 21 for the old generation and the new generation. This is an example of the occurrence of 1 in a graph, and shows three typical states related to the garbage collection 210 and newGC 211 occurrence intervals and memory usage (201, 202).

In the graphs in Figs. 8 (A) to 8 (C), the vertical axis represents the amount of memory, and the horizontal axis represents time. In Figs. 8 (A) to (C), the dotted line indicates the set maximum heap size 220, and the dashed line indicates the boundary 221 between the new generation and old generation memory areas. The area below the dashed line is the area of the old generation memory, and the area from the dashed line to the dotted line (maximum heap size 220) is the area of the new generation memory. The solid line in each generation shows the temporal change in memory usage (201, 202) for each generation. The states in Fig. 8 (A) to (C) are shown in Figs. 6 (A) to (C). Corresponding to each state.

When the new generation's memory usage 202 exceeds the new generation's memory due to a memory allocation event (in FIG. 8, when the maximum heap size exceeds 220), the new VM 1 Execute 1 to perform a burst of new generation memory. In addition, the Java virtual machine 1 transfers objects remaining in the new-generation memory to the old-generation memory after experiencing the new wGC 211 several times. Performs garbage collection 210 of the entire memory when objects cannot be moved from

The frequency of occurrence of new GC211 is higher than the frequency of occurrence of GC210 in the entire memory. However, while the processing time of the GC 210 is several seconds, and in some cases ten and several seconds, the processing time of the new GC 211 is about several tens of milliseconds. In applications where response time is important, it is necessary to maintain the state where GC210 does not occur as much as possible (the state shown in Fig. 8 (A)).

GC210 occurs when there are many objects that move from the new generation to the old generation, and when there are many long-lived objects and when the new generation has a small memory size and overflows ( Even if the generation is young, it may be moved to the old generation (when overflow algorithm is adopted). At the time of the new GC 211, the size of the objects overflowing from the new generation memory to the old generation memory (hereinafter referred to as the overflow amount) was measured, and the state shown in Figs. 8 (B) and 8 (C) was obtained. The amount of memory required to achieve the state shown in Fig. 8 (A) can be estimated.

Fig. 9 is a diagram explaining the measurement of the overflow amount. Figure 9 (A) shows the state in which objects overflow from the memory of the new generation 240 to the memory of the old generation 230 in the new GC 211, where the hatched area indicates the overflow amount. Here, it is assumed that the amount of memory of the new generation 240 was sufficient at the time when the newGC 211 occurred (the timing of the occurrence of the newGC 211 has been extended). It shall be expressed as shown in Fig. (B). If this is repeated for every new GC 2 11 1, the state will be as shown in Fig. 9 (C).

Figure 9 (C) shows a state where it is assumed that the object that actually overflowed with four new GCs 211 did not overflow with an infinite amount of memory of the new generation 240. are doing. When the k-th overflow amount and expressed as t _k, overflowing total amount of which is assumed to overflowing Do bought at the time of FIG. 9 (C) is a + t ₂ + t + t. If there is an extra memory of t + t ₂ + t + t ₄ in addition to the original memory in the area of the new generation 240 from the beginning, the new generation can be performed up to four times with new GC 2 11 No objects overflow from the 240 memory.

In addition, some objects become unnecessary over time, and the amount of overflow is estimated as follows. For example, in Fig. 9 (C), suppose that the object overflowed by the first newGC 211 does not overflow and remains in the area of the new generation 240 until the fourth newGC 211 time. Some of them should be unnecessary objects. Similarly, among the objects that overflowed at the time of the second ne wGC 21 1 and the objects that overflowed at the time of the third ne wGC 21 1 You. In Fig. 9 (C), some of the objects that were determined not to have overflowed by the fourth new GC 211 may be released by the fourth new GC 211. Actually, Fig. 9 (C) is considered to be the state shown in Fig. 9 (D).

In FIG. 9 (D), when the one ne WGC 2 1 1 by viability object d, the k-th overflow amount and expressed as x _k,

X! = t! X d ³

= t X d ²

xa = t X d ¹

x = t X d ⁰

And the total overflow is X + χ ₂ + χ ₃ + 4. Therefore, the new generation 24

If the amount of memory 0 is extra only _{_{X l + X 2 + X 3}} + X 4, so that the has failed cause overflow.

In general, the cumulative amount of overflow of n new GCs can be obtained by the following equation (3). In equation (3), x _t . _tal indicates the total overflow amount.

n (t _k X d ⁿ - ^k ) · · Equation (3) Hereafter, measurement of memory usage · Analysis of generation-type garbage collection by means 11 Shows the magic of determining the danger of lack of memory according to the features. The judgment of the danger of memory shortage in the generational garbage collection is obtained by adding the following (Judgment 4) and (Judgment 5) to (Judgment 1) to (Judgment 3) described above.

Note that in the following generation-type garbage collection judgment, the following equation (4) is used instead of equation (1) for calculating the memory usage rate.

Usage rate = Old generation memory usage Z Old generation capacity · · · Formula (4)

(Decision 4) The memory usage after garbage collection is measured, and even if the memory usage is below the memory shortage warning threshold value of 106 (even if the memory is fully recovered), GC occurs. If the interval is less than the GC occurrence interval warning threshold value 103 and the ratio of the new generation memory amount (the new generation capacity total capacity) is less than the new generation shortage warning threshold value 108, the new generation Judgment that the memory size is too small affects the performance (increases the frequency of occurrence of GC).

(Judgment 5) The total overflow amount is calculated after every new GC by equation (3), and if (total overflow amount / free space of old generation) exceeds the overflow warning threshold value 109, the OutOf Memory Is determined to be at risk.

The memory usage measurement / analysis means 11 matches the basic judgment results shown in (Judgment 1) to (Judgment 3) and the characteristics of the generational garbage collection shown in (Judgment 4) and (Judgment 5). Based on the result of the risk judgment, a warning of the risk of OutOfMemory and a warning of the risk of performance degradation due to insufficient memory are issued.

In addition, the memory usage measurement 'analysis means 11 calculates the amount of memory required to recover the state. For example, the memory size required for recovery in (Judgment 1) to (Judgment 3) is such that the vacancy rate of memory after garbage collection (vacancy rate = 1 usage rate) exceeds the memory vacancy recovery threshold value of 107 Is the size calculated in The calculation is to find m (capacity added for recovery) that satisfies the following equation (5).

(Old generation free space + m) / (old generation space + m)

> Memory free space recovery threshold 1 0 7 ·. '(5) For example, in the case of (Judgment 5), the amount of additional memory required for recovery is determined by the amount of memory between one GC and the next GC. For the generated new GC, let it be the total amount of overflow determined by Eq. (3). As shown in Fig. 8, the ratio between the area of the new generation memory and the area of the old generation memory in the generation-type garbage collection can be set as an option when the JaVa virtual machine 1 is started. is there. In addition, in the generational garbage collection as shown in Fig. 8, the memory size of the new generation and the memory size of the old generation are kept constant, but the memory size of the new generation and the memory size of the old generation are dynamically changed. There is also a generational garbage collection.

FIG. 10 is a diagram for explaining the state of the generation interval of garbage collection and the memory usage in the generation type garbage collection that dynamically changes the size. In the generation-type garbage collection that dynamically changes the size shown in Fig. 10, the free space of the memory immediately after garbage collection 210 is reduced by a preset ratio between the new generation memory area and the old memory area. It is distributed to the generation memory area.

Figures 10 (A) to (C) show changes in memory usage (old generation 201, new generation 202) and garbage collection 210, neGC 211 for the old generation and the new generation, respectively. Fig. 10 (A) shows a typical example of the state of garbage collection 210, new GC 211, and the memory usage. In the graphs (C) to (C), the vertical axis represents the amount of memory, and the horizontal axis represents time. In Figs. 10 (A) to 10 (C), the dotted line indicates the set maximum heap size 220 and the dashed line indicates the boundary 221 between the new generation and old generation memory areas. The area below the dashed line is the area of the old generation memory, and the area from the dashed line to the dotted line (maximum heap size 220) is the area of the new generation memory. The solid line in each generation indicates the time variation of the memory usage (201, 202) for each generation. The states shown in FIGS. 10 (A) to (C) correspond to the states shown in FIGS. 6 (A) to (C).

When the memory allocation event causes the new generation memory usage 202 to exceed the new generation memory amount (in FIG. 10 when the maximum heap size 220 exceeds the maximum heap size 220), the J a Va virtual machine 1 Execute new GC211 to release the new generation memory. In addition, Java virtual machine 1 is a new generation after experiencing new GC 2 11 several times. Move the objects remaining in the memory of the old generation to the memory of the old generation. The memory of the new generation is reduced and the memory of the old generation is increased by the amount of moving objects from the memory of the new generation to the memory of the old generation.

When the memory of the old generation becomes full, & 3 virtual machine 1 executes garbage collection 210 of the entire memory. At this time, the free space of the memory immediately after garbage collection 210 is distributed to a new generation memory area and an old generation memory area at a preset ratio.

Memory usage measurement · The judgment of the risk of memory shortage in generational garbage collection by dynamically changing the size by the analysis means 11 is (Judgment 1) to (Judgment 3) and (Judgment 5) described above. Equation (1) is used to calculate the usage rate in generational garbage collection that dynamically changes size. The memory usage measurement and analysis means 11 is based on the risk judgment results shown in (Judgment 1) to (Judgment 3) and (Judgment 5). Give a warning of drop.

The memory usage measurement and analysis means 11 calculates the amount of memory required to recover the state. In generational garbage collection that dynamically changes size, the amount of memory required for recovery is improved by the above equation (2). For example,

The memory size required for recovery in (Judgment 1) to (Judgment 3) was calculated so that the ratio of the free space of the old generation to the total capacity after garbage collection exceeded the memory free space recovery threshold value of 107. Size. The calculation is to find m (capacity added for recovery) that satisfies the following equation (6).

(Old generation free space + old generation distribution ratio X m) / (total space + m)

> Memory free recovery threshold 1 0 7 ··· 'Eq. (6) Also, the amount of additional memory required for recovery in case (judgement 5) is the same as in the case of the generational garbage collection described above. .

An example of processing in the memory usage measurement and analysis means 11 will be described with reference to the flowcharts in FIGS.

In the flowcharts in Figs. 11 to 15, "non-dynamic generation type GC" means a generation type GC that does not dynamically change the memory size of the new generation and the old generation. Indicates a page collection. "Dynamic generation type GC" indicates a generation type garbage collection that dynamically changes its size. "GC" indicates the garbage collection of the entire memory, and "newGC" indicates the new generation garbage collection in the generation type GC.

Fig. 11 is a flowchart of the analysis process of memory usage measurement.

First, it is determined whether or not the memory usage measurement flag 101 is ON (step S1), and if not (if OFF), the processing is terminated. If it is ON in step S1, it is determined whether or not GC has occurred (step S2). If GC has not occurred, the process proceeds to step S6.

If a GC occurs in step S2, the data measurement and calculation processing shown in Fig. 12 is performed (step S3), and the data analysis and warning processing A shown in Fig. 13 are performed (step S4). ), Then perform data analysis and warning processing B shown in Fig. 14 (step S5), and proceed to step S6.

It is determined whether or not it is a generation type GC (step S6), and if it is not a generation type GC, the process is terminated.

If it is a generation type GC in step S6, the generation type GC specific measurement-analysis processing shown in Fig. 15 is performed (step S7), and the processing ends.

Figure 12 is a flowchart of the data measurement and calculation process.

First, the current GC occurrence time is measured (step S10), and the current GC occurrence interval (the current GC occurrence time minus the previous GC occurrence time) is calculated (step S11). Next, the memory usage and the free space immediately after the GC are measured (step S12), and the memory usage and the free space immediately after the GC (free space = 1-one usage) are calculated (step S1). 3), End the processing.

Here, in step S12, in the case of the generation type GC, measurement is performed not only on the entire memory but also on the memory of the old generation. In addition, in the calculation of the memory usage rate in step S13, when using non-generational type GC and dynamic generation type GC, calculation is performed using equation (1). Is calculated using equation (4).

Figure 13 is a flowchart of data analysis and warning processing A. First, it is determined whether or not “this usage rate ≥ th 2 (OutOfMemory warning threshold A 104)” is satisfied (step S20), and if it is, the process proceeds to step S26. If not satisfied in step S20, it is determined whether or not “current usage rate> previous usage rate” is satisfied (step S21). If not, how many times the memory usage rate increases. The counter c1 for counting whether or not it is continuous is set to 0 (step S23), and the process ends.

If it is satisfied in step S21, it is determined whether or not “this usage rate> th3 (OutOfMemory warning threshold B105)” is satisfied (step S22). If not, use of memory is determined. The counter c1 for counting how many times the rate increase continues is set to 0 (step S23), and the process ends.

If the condition is satisfied in step S22, the counter c1 is incremented by 1 (c1 = c1 + 1) (step S24).

It is determined whether or not the value of the counter c1 is equal to or greater than a value n1 arbitrarily set in advance (step S25). If the value is equal to or greater than n1, the process proceeds to step S26. If the value is not equal to or more than the value n1 in step S25, the process ends.

In step S26, a warning of the danger of OutOfMemory (warning w1) is issued, the amount of memory required for recovery is calculated (step S27), and the process ends.

In the calculation of the amount of memory required for the recovery in step S27, if the generation type GC is not used, a calculation is performed to find m (the capacity added for recovery) that satisfies the above equation (2), In the case of a type GC, a calculation is performed to find m (capacity added for recovery) that satisfies the above equation (5). Calculation to determine the capacity to be added).

Figure 14 is a flowchart of the data analysis / warning process B.

First, it is determined whether or not “this GC generation interval <th 1 (GC generation interval warning threshold value 103)” is satisfied (step S30), and if not, how many times the GC generation interval continues. The counter c 2, which counts whether the value has fallen below the GC generation interval warning threshold value 103, is set to 0 (step S 31), and the process ends.

If the condition is satisfied in step S30, the counter c2 is incremented by 1 (c2 = c2 + l) (step S32).

It is determined whether or not the value of the counter c2 is equal to or greater than a value n2 arbitrarily set in advance (step S33), and if not, the process is terminated.

In step S33, if c2 is equal to or larger than value n2, it is determined whether or not "this usage rate ≥ th4 (memory shortage warning threshold value 1 06)" is satisfied (step S34). If the conditions hold, a warning (warning w2) is issued to warn of performance degradation due to insufficient memory (step S35), the amount of memory required for recovery is calculated (step S36), and the process ends.

In the calculation of the amount of memory required for the recovery in step S36, if the generation type GC is not used, a calculation for finding m (the capacity added for recovery) that satisfies the above equation (2) is performed. In the case of a GC, calculation is performed to find m (capacity to be added for recovery) that satisfies the above equation (5). To calculate).

If not satisfied in step S34, it is determined whether or not "vacancy rate ≥ th5 (memory free space recovery threshold value 1 07)" (step S37). A warning (warning w3) is generated (step S38), and the process ends.

If the condition is satisfied in step S37, it is determined whether the GC is a non-dynamic generation type GC (step S39). If not, the absolute amount of memory is smaller than the memory consumption speed. A warning (warning w4) is issued (step S40), and the process ends.

If it is a non-dynamic generation type GC in step S39, it is determined whether "new generation capacity Z total capacity <th6 (new generation shortage warning threshold value 1 08)" is satisfied (step S4). 1) If this is true, a warning is issued (warning w5) that the proportion of the new generation is small (step S42), and the processing ends. .

If this is not the case in step S41, a warning (warning w6) is issued (warning w6) indicating that the absolute amount of memory is small compared to the memory consumption speed, and the ratio of the survivor capacity is small (step S43), and the processing ends. I do. Figure 15 is a flowchart of the generation-type GC-specific measurement and analysis process.

First, it is determined whether or not newGC has occurred (step S50), and if not, the process ends.

If a new GC has occurred in step S50, it is determined whether or not the new GC is the first new GC after the GC (step S51). If the new GC is not the first new GC, the process proceeds to step S53. If it is the first new GC in step S51, the value n is set to 1 (step S52), and the process proceeds to step S53.

The amount of objects overflowing from the new generation to the old generation is measured by new GC (step S53), and the total amount of overflow is calculated using equation (3) (step S54). Next, it is determined whether or not “total overflow / free space of the previous generation> th 7 (overflow warning threshold value 1 09)” is satisfied (step S 55). Proceed to.

If the condition is satisfied in step S55, a warning of the danger of OutOfllemory (warning w1) is issued (step S56), and the total overflow amount calculated in step S54 is set as the amount of memory to be added for recovery ( Step S57), and proceed to step S58. The value n is incremented by 1 (n = n + l) (step S58), and the process ends.

Figure 16 is a diagram showing an example (1) of a state in which the memory usage measurement and analysis means is subject to a warning. Figs. 16 (A) and (B) correspond to the judgment conditions in Fig. 13 respectively. In the graphs in Figs. 16 (A) and (B), the vertical axis shows the amount of memory and the horizontal axis shows time.

In Fig. 16 (A), the bold line 104 'is the value obtained from (full capacity XOutOfMemory warning threshold A104). In Fig. 16 (B), the bold line 105 'is the value obtained from (the total capacity XOutOfMemory warning threshold B105). In Figs. 16 (A) and (B), the thick line 107 'is the value obtained by (total capacity X (1-memory free recovery threshold value 107)).

FIG. 16 (A) shows an example in which step S20 in FIG. 13 is established. Looking at the change in memory usage 200 in Fig. 16 (A), almost no free memory has been recovered by GC210. Memory usage after GC 210 0 200 Indicates that the memory usage exceeds the OutOfMemory warning threshold A104. The OutOfMemory warning threshold A104 is, for example, about 90% to 95%.

Fig. 16 (B) is an example in which step S21, step S22, and step S25 in Fig. 13 are simultaneously established. In Fig. 16 (B), looking at the change in the memory usage 200, the free space of the memory gradually stops recovering in the order of GC210o; GC210i3 and GC210y. You can see that there is. In addition, since the memory usage 200 after GC 210; GC and GC 210 g continuously exceeds the thick line 105 ', the memory usage after GC 210 j3 and GC 210 y, respectively. It can be seen that the memory usage exceeds the 0utOfMemory warning threshold B105.

Figure 17 is a diagram showing an example (2) of a state in which the memory usage measurement and analysis means is subject to a warning. FIGS. 17 (A) to 17 (C) correspond to the judgment conditions in FIG. 14, respectively. In the graphs in Figs. 17 (A) to 17 (C), the vertical axis represents the amount of memory, and the horizontal axis represents time.

In Fig. 17 (A), the bold line 106 'is the value determined by (total capacity X memory shortage warning threshold value 106). In Fig. 17 (C), the bold line 108 'is the value obtained by (total capacity X (1-new generation shortage warning threshold value 108)). In Figs. 17 (A) to 17 (C), the thick line 107 'is the value obtained by (total capacity X (1-memory free recovery threshold value 107)).

FIG. 17 (A) shows an example in which step S34 in FIG. 14 is established. Looking at the change in memory usage 200 in Fig. 17 (A), the memory space is recovered to some extent by GC210, but the GC generation interval is shortened. Since the memory usage 200 after GC210 exceeds the thick line 106 ', it can be seen that the memory usage exceeds the memory shortage warning threshold value 106. The memory shortage warning threshold value 106 is, for example, about 80% to 90%.

Fig. 17 (B) is an example where step S37 in Fig. 14 is satisfied, and is an example where the generation type GC is not used. Looking at the change in the memory usage 200 in Fig. 17 (B), the memory is recovered by GC210. In addition, since the memory usage 200 after GC 210 is below the thick line 107 ', It can be seen that the vacancy rate exceeds the memory vacancy recovery threshold value 107.

However, the GC occurrence interval is becoming shorter. This is because the absolute amount of memory (maximum heap size 220) is small. Memory free recovery threshold 1 0 7

, For example, about 50% and 60%.

Fig. 17 (C) is an example in which step S41 in Fig. 14 is satisfied, and is an example in the case of non-dynamic generation type GC. In Fig. 17 (C), a look at the change in the memory usage 201 of the old generation shows that the free memory has been recovered by GC210. In addition, the memory usage rate of the old generation after GC 210 is lower than the bold line 107 '. I have.

Therefore, the GC generation interval is shortened. In addition, since the boundary between the old generation memory and the new generation memory 221 exceeds the thick line 108 ', the ratio of the new generation capacity to the total memory capacity is the new generation shortage warning threshold. It turns out that it is less than 108. This is because the new generation capacity is small. The new generation shortage warning threshold value 1108 is, for example, about 5% to 10%.

Next, the thread measurement / analysis means 12 that measures and analyzes whether or not the thread of the JaVa application has a deadlock and issues a warning is described.

The state of the thread of JaVa includes a state of being executed, a state of being ready for execution, a state of being suspended, a state of sleeping, and a state of waiting for lock release.

As described in the fourth prior art, the thread state transition can be measured by events of the Java virtual machine / profiler interface. In fact, there are products that detect these dead events by analyzing these events and examining the thread status.

However, thread state transitions occur very frequently, and if all these events are measured, a heavy load will be placed on application operation. Therefore, deadlock detection, which measures and analyzes thread state transitions based on events in the Java Virtual Machine 'profiler interface, should be performed during development.

In contrast, in this system, the thread measurement and analysis Measure the thread status periodically at intervals that do not impose a load on the operation of the thread. In addition, the thread that is waiting for lock release is recorded, and it is checked whether the same thread is waiting for relock release at the next measurement. If the machine is standing by continuously, count the number of times. If there are two or more threads that have exceeded the deadlock warning threshold value 110 for the number of consecutive wait states for lock release, it is determined that there is a risk of deadlock.

This makes it possible to detect the risk of deadlock during application operation without imposing a load on the operation.

Figure 18 is a flowchart of the thread measurement and analysis process.

First, it is determined whether or not the thread measurement flag 102 is ON (step S60). If it is not ON (if it is OFF), the process proceeds to step S68.

If it is ON in step S60, the state of all threads is measured (step S61), and the threads that are currently waiting for lock release are extracted and recorded (step S62). In addition, among the threads that have already been recorded, the threads that have not been extracted this time are deleted (step S63).

The counter of the number of times that the thread waiting state extracted for the first time this time is continuous (this is called a continuous counter) is set to 1 (step S64), and the continuous counter of the thread extracted last time is incremented by 1 (step S65). ).

Next, it is determined whether or not there are two or more threads satisfying “continuous counter ≥ th 8 (deadlock warning threshold 1 10)” (step S66). Proceed to 6.

If it exists in step S66, a warning (warning w7) that there is a danger of deadlock is issued (step S67), and the process proceeds to step S68.

Sleeps the thread state measurement at the measurement interval specified as an option when the JaVa virtual machine 1 is started or as a command from the JaVa virtual machine control means 20 (described later) of the application server 2. Then (step S68), the process returns to step S60.

As described above, this system employs an algorithm that makes judgments based on periodic measurements, rather than measuring all thread state transitions. The risk of thread deadlock can be warned without imposing a load on the operation environment of the location.

Next, we explain the technical means that the Java virtual machine 1 notifies the analysis result to the upper layer (application server 2) and feeds it back so that the optimal operation environment is established.

Fig. 19 is a diagram showing the distribution of processing to the Java virtual machine by cooperation between the Java virtual machine and the application server. In the system shown in Fig. 19, there are four Ja and Va virtual machines 1a to 1d in the upper layer of OS4, each of which has measurement and analysis means 10a to 10d. An application server 2 is located above the JaVa virtual machines 1a to 1d.

Each of the JaVa virtual machines 1a to 1d is started by the upper-level application server 2, and there is a close connection between each of the JaVa virtual machines 1a to 1d and the application server 2. Have a relationship. The application server 2 starts a plurality of Java virtual machines 1a to 1d, distributes requests from clients to each of the Java virtual machines 1a to 1d, and processes the application.

The results of the analysis performed by the measurement and analysis means 10a to 10d in each JaVa virtual machine 1a to 1d are reported directly to the application server 2. The application server 2 feeds the received analysis result notification to the application operating environment and distributes requests from clients to the Java Virtual Machines 1a to 1d more appropriately.

For example, it is assumed that the Java virtual machine 1a sends a notification to the application server 2 that there is a high risk of an OutOfMemory error due to insufficient memory. The application server 2 stops distributing the processing to the Java virtual machine 1a, and when the virtual machine 1a completes the current processing, the new JaV a. aVirtual machine 1 Processing can be distributed to b to ld.

First, the cooperation between the Java Virtual Machine 1 and the application server 2 will be described.

Figure 20 is a diagram showing the cooperation between the Java virtual machine and the application server. The J a V a virtual machine control means 20 in the application server 2 This is a means to start and stop the ava virtual machine 1 and to distribute application processing to the Java virtual machine 1. The Java virtual machine control means 20 and the Java virtual machine 1 are linked to each other by the Java virtual machine link means 3.

The JaVa virtual machine cooperation means 3 includes an analysis result notifying means 31. The analysis result notification means 31 has a function of notifying the analysis result directly by inter-process communication and a function of notifying the analysis result via a medium such as a file.

The measurement and analysis means 10 in the Java virtual machine 1 sends the analysis result to the Java virtual machine control means 20 using the analysis result notifying means 31.

The Java virtual machine control means 20 controls the start and stop of the Java virtual machine 1 and the Java virtual machine based on the analysis result sent by the analysis result notification means 31. Distribution of the application to the application. Here, the Java virtual machine control means 20 starts and stops the Java virtual machine 1 and distributes applications to the Java virtual machine 1 based on information from the Java virtual machine 1. The means for optimally performing is Java virtual machine optimal control means 21.

Further, the Java virtual machine cooperation means 3 includes a command notification means 32. This command notification means 32 notifies a command directly by inter-process communication.

The JaVa virtual machine control means 20 uses the command notification means 32 to set commands for instructing start / stop of measurement / analysis for each type of measurement, and setting of thresholds used for judgment of analysis results Send commands to change, commands to force garbage collection, etc. For example, if a warning is not notified for a long period of time, the JaVa virtual machine control means 20 uses the command notification means 32 to notify the JaVa virtual machine 1 of memory usage. A command can be sent to instruct to suspend the measurement.

FIG. 21 is an analysis result notification processing flowchart. Here, the analysis result notification processing is explained separately for the processing of the Java virtual machine 1 'processing on the analysis means 10 side and the processing on the Java virtual machine control means 20 side of the application server 2. Fig. 21 (A) is a flowchart of the analysis result notification processing on the measurement and analysis means 10 side. It is. The measurement / analysis means 10 determines whether or not there is a warning (step S70), and if there is a warning, notifies the Java virtual machine control means 20 using the analysis result notifying means 31. (Step S71). At this time, if there is information on the amount of memory required for recovery, it is also notified. The warning may be sent by inter-process communication or output to a log file.

FIG. 21 (B) is an analysis result notification processing flow chart on the Java virtual machine control means 20 side. J _a V _a virtual machine control unit 20 determines whether it has received a warning (step S 80), if receiving a warning, pass the warning to the J ava virtual machine optimal control unit 2 1 (Step S81). At this time, if information on the amount of memory required for recovery has been received, it is also passed.

FIG. 22 is a flowchart of the command notification process. Here, the command notification process is described separately for the process on the Java virtual machine control means 20 side of the application server 2 and the process on the measurement and analysis means 10 side of the Java virtual machine 1. (A) is a flowchart of the command notification processing on the Java virtual machine control means 20 side. The _Java virtual machine control means ₂₀ determines whether or not there is a command to change the operating environment to the measurement and analysis means 10 (step S90). If there is a command, the command notification means _{3.2 is issued} . A command to change the setting is sent to the measurement / analysis means 10 using this (step S91). Commands are sent by inter-process communication.

Here, examples of commands for changing the operating environment include the following.

(1) Memory usage measurement start command

(2) Memory usage measurement stop command

(3) Thread measurement start command

(4) Thread measurement end command

(5) Threshold setting change command

(6) Force garbage collection command

Fig. 22 (Β) is a flowchart of the command notification processing on the measurement / analysis means 10 side. It is. The measurement 'analyzing means 10 determines whether or not a command has been received (step S100), and if the command has been received, executes the command (step S101). '

Here, examples of command execution include the following.

(1) Start memory usage measurement If it is a Z suspend command, set the memory usage measurement flag 101 to ONZ〇FF.

(2) Thread measurement start If the command is a Z end command, the thread measurement flag 102 is set to 〇N / 0 FF.

(3) If the command is a threshold change command, set / change the corresponding threshold (103 to 110).

(4) If it is a forced garbage collection command,] & & Call the garbage collection process of virtual machine 1.

Next, the feedback of the analysis results to the application operating environment is described.

Measurement in Java virtual machine 1. As a result of analysis by analysis means 10, notifications such as warning of danger of OutOfMemory, warning of performance degradation due to insufficient memory, and memory size required for recovery, etc. It is sent to the virtual machine control means 20. Here, assuming that the Java virtual machine 1 and the Java virtual machine control means 20 are directly connected to each other by force inter-process communication, the notification reaches the application server 2 immediately. The Java virtual machine control means 20 periodically checks whether or not the notification has been received, and, upon receiving the notification, immediately passes the notification to the Java virtual machine optimum control means 21 to execute the application operating environment. Feed pack. For example, when receiving a notification of a danger of OutOfMemory from the Java virtual machine 1, the Java virtual machine optimal control means 21 stops distribution of application processing to the Java virtual machine 1. The J a V a After the virtual machine 1 is completed applications that are currently being processed, launch the alternative of J _a V _a virtual machine 1 '. When launching the alternative Java virtual machine 1 ′, the startup options of the alternative Java virtual machine 1, based on the memory size required for recovery notified from the Java virtual machine 1, are used. Specify the optimal memory size. Also, when a Java virtual machine 1 receives a warning of performance degradation due to lack of memory, the Java virtual machine optimal control means 21 is notified the next time the Java virtual machine 1 starts up. Based on the memory size required for recovery, specify the optimal memory size in the Java virtual machine 1 startup options. There is also a feed pack method that reduces the distribution of application processing to the Java virtual machine 1 that warns of performance degradation due to insufficient memory.

The notification of the thread deadlock danger warning is not fed back to the startup of the Java virtual machine 1 or the distribution control of application processing. This information is left in the file as log information of the application server 2 and used for later review of the application.

FIG. 23 is a flowchart of the JaVa virtual machine optimal control process.

First, the memory size required for warning and recovery is received (step S110), and the type of warning is determined (step S111).

If the warning of the danger of OutOfMemory (warning _w1 ) is found in step _{S111, the Java} virtual machine 1 that has issued the warning is terminated after the application is terminated (step _S112 ). The alternative Java virtual machine 1 'is started (step S113), and the processing ends. When the alternative Java virtual machine 1 is started, the optimal memory size is specified as an option based on the memory size required for the received recovery.

If a warning of performance degradation due to insufficient memory (warning w2) is issued in step S111, the received memory size required for recovery is saved (step S114), and the process ends. At the next startup of the Java virtual machine 1, the optimal memory size is specified in the options based on the saved memory size required for recovery.

If the warning of the risk of deadlock is found in step S111 (warning w7), a warning of the risk of deadlock is recorded in the log (step S115). This is notified to the operation manager (step S116), and the processing ends.

The effects of this system are as follows. In this system, instead of measuring all of the memory allocation and release, the judgment is made based on the measurement when the garbage collection event occurs and the periodic measurement of the memory usage as necessary. Since it uses the algorithm, without placing a load on the application of the production environment, 0ut0: Ru can be a warning of performance degradation due to the risk and out-of-memory fMem _O ry.

In the method of collecting measured data and analyzing the data by an external process, it takes time from data measurement to feed pack for recovery, which may impair the time consistency. In this system, there is a means to measure and analyze data in the Java virtual machine running the application, so that a more timely feed pack can be performed.

In addition, this system has multiple stages of improved algorithms for judging danger according to the various garbage collections used by each JaVa virtual machine, so that more accurate danger judgment can be performed. Can be.

In addition, a simple danger of OutOemory or a warning of performance degradation due to insufficient memory may cause a similar situation again, but this system is required for optimal operation of the JaVa virtual machine. Since a notification is also made when the amount of memory is large, more appropriate feedback for recovery can be provided.

In this system, the means of measuring and analyzing the data of the Java virtual machine and notifying the danger and the means of controlling the Java virtual machine of the upper layer (application server) are directly communicated by interprocess communication. Since they are connected and can automatically issue warnings and provide feedback without human intervention, the risk of delaying recovery from a problem to recovery is reduced. Industrial applicability

The present invention monitors the state of the execution environment of the middle tier, which is located on the server side, during the operation of an application in a three-tier system consisting of the client tier, middle tier, and database tier that constitute an enterprise application. By avoiding the state, it can be used to operate the application stably. As an execution environment of the middle tier deployed on the server side, for example, there is an execution system such as a Java virtual machine having a memory management function by garbage collection. In the present invention, the means for measuring and analyzing data such as the JaVa virtual machine and notifying the danger and the means for controlling the JaVa virtual machine in the upper layer (application server) are linked with each other to generate a warning. Notifications and feed packs can be performed automatically without manual intervention, so that the risk of memory shortage in the execution environment can be efficiently avoided.

Claims

The scope of the claims

1. A method of estimating and avoiding danger of an execution environment in a computer system having a program execution environment in which allocated memory is released by garbage collection.

Measuring the memory usage immediately after garbage collection when a garbage collection event occurs;

Estimating the risk of memory shortage from the measured memory usage, and calculating the amount of memory required to avoid the predicted danger of memory shortage;

A warning of the predicted danger of memory shortage and a step of notifying another application of an amount of memory required to avoid the warning.

Danger prediction of execution environment characterized by the following.

2. Predict the risk of memory shortage and calculate the amount of memory required to avoid it using algorithms that correspond to the types of garbage collection used by the system.

The danger prediction / avoidance method for the execution environment according to claim 1, characterized in that:

3. The process of periodically measuring the thread status and extracting the thread in the waiting state while waiting for the release of the lock;

Counting the number of consecutive extractions for each thread in the waiting state waiting for release of the extracted lock;

If there are two or more threads whose number of consecutive extractions exceeds a predetermined threshold, it is determined that there is a risk of deadlock and a warning is output.

4. Danger prediction / avoidance method of the execution environment in a computer system having a program execution environment in which the allocated memory is released by garbage collection,

When garbage collection occurs, measure the interval between garbage collections. Process,

When the measured garbage collection interval is continuously lower than a predetermined threshold, the process of determining that the performance is degraded due to insufficient memory,

Notifying other applications of a warning that performance is degraded due to insufficient memory.

A danger prediction / avoidance method for an execution environment, characterized in that:

5. Danger prediction / avoidance method for the execution environment in a computer system having a program execution environment in which the allocated memory is released by garbage collection,

Measuring the amount of memory used immediately after garbage collection when a garbage collection event occurs;

Determining that there is a risk of a memory shortage error when the measured ratio of the memory usage to the total memory capacity exceeds a predetermined threshold value;

Notifying another application of a warning that there is a risk of an insufficient memory error.

A method for avoiding danger prediction in an execution environment, characterized in that:

6. Danger prediction of the execution environment in a computer system that has a program execution environment that releases allocated memory by garbage collection.

When a garbage collection event occurs, immediately after the garbage collection (^ the process of measuring memory usage,

If the ratio of the measured memory usage to the total memory capacity tends to increase continuously and continuously exceeds a predetermined threshold, there is a risk of a memory shortage error. And the process of determining

Danger prediction of execution environment characterized by the following.

7. A method for avoiding danger prediction of an execution environment in a computer system having a program execution environment in which allocated memory is released by generational garbage collection,

When the event of garbage collection of the entire memory occurs, the process of measuring the memory usage immediately after the garbage collection of the entire memory and the occurrence interval of the garbage collection of the entire memory;

The amount of memory allocated to the new generation is too small if the measured garbage collection interval of the entire memory is less than a predetermined threshold because the amount of memory allocated to the new generation is small. And the process of notifying another application of a warning that the amount of memory allocated to the new generation is too small.

8. A danger prediction / avoidance method for an execution environment in a computer system having a program execution environment in which allocated memory is released by generational garbage collection,

Measuring the amount of overflow from the new generation area to the old generation area when a garbage collection event occurs in the new generation area;

If the sum of the overflow calculated based on the measured overflow to the free space in the old generation area exceeds a predetermined threshold, the risk of memory shortage error is increased. The process of determining that there is,

9. In a computer system with a program execution environment that releases allocated memory by garbage collection,

Means for measuring memory usage immediately after garbage collection when a garbage collection event occurs;

Means for predicting the risk of memory shortage from the measured memory usage, and calculating the amount of memory required to avoid the predicted risk of memory shortage Means,

A warning is provided for the predicted danger of memory shortage and means for notifying another application of the amount of memory required to avoid the warning is provided.

Danger prediction of execution environment characterized by the following.

10. Program for releasing allocated memory by garbage collection A program for predicting and avoiding danger of the execution environment in a computer system having the execution environment.

When the event of garbage collection occurs, the process of measuring the memory usage immediately after garbage collection;

A process of predicting the risk of memory shortage from the measured memory usage, and a process of calculating a memory amount required to avoid the predicted risk of memory shortage;

The warning about the predicted danger of insufficient memory and the process of notifying another application of the amount of memory necessary to avoid the warning are as follows.

An execution environment danger prediction / avoidance program to be executed by a computer.

11. A program for releasing allocated memory by garbage collection A computer-readable recording medium storing a program for predicting and avoiding danger of an execution environment in a computer system having an execution environment,

A process of notifying another application of the warning about the predicted danger of lack of memory and the amount of memory required to avoid the warning,

Recorded program to be executed by computer

Danger prediction of execution environment characterized by the following: A recording medium for Z avoidance program.